Real-Time Traffic Sign Detection and Recognition Using Faster R-CNN with VGG16 Backbone

Real-time detection and recognition of traffic signs is an inevitable part of advanced driver assistance systems (ADAS) and autonomous driving. Here in this paper, we come up with a traffic sign detection system which employs Faster R-CNN as its framework and VGG16 for extracting features. This system successfully implements traffic sign detection and classification with high accuracy and speed even in unfavorable conditions like illumination changes, occlusion, and complex background. The Region Proposal Network successfully filters down the possible regions of interest, and the VGG16 as a backbone, extracts more features useful in traffic sign detection. In order to justify the proposed model, we use treat public databases such as the German Traffic Sign Detection Benchmark (GTSDB). The test set results of the proposed system are a precision of 93.4, a recall of 92.1, an F1-score of 92.7, and an mAP of 94.5. In addition, optimizations are included for real-time functionality with mean inference time equal on average 88 ms. Comparing it with other models such as YOLOv3 and SSD shows that the proposed system is more accurate than it is faster. Based on the obtained results, it can be argued that the described approach is perfect for real traffic sign detection in ITS.